xref: /freebsd/sys/netpfil/pf/pf_norm.c (revision 389e4940069316fe667ffa263fa7d6390d0a960f)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright 2001 Niels Provos <provos@citi.umich.edu>
5  * Copyright 2011 Alexander Bluhm <bluhm@openbsd.org>
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27  *
28  *	$OpenBSD: pf_norm.c,v 1.114 2009/01/29 14:11:45 henning Exp $
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include "opt_inet.h"
35 #include "opt_inet6.h"
36 #include "opt_pf.h"
37 
38 #include <sys/param.h>
39 #include <sys/kernel.h>
40 #include <sys/lock.h>
41 #include <sys/mbuf.h>
42 #include <sys/mutex.h>
43 #include <sys/refcount.h>
44 #include <sys/socket.h>
45 
46 #include <net/if.h>
47 #include <net/vnet.h>
48 #include <net/pfvar.h>
49 #include <net/if_pflog.h>
50 
51 #include <netinet/in.h>
52 #include <netinet/ip.h>
53 #include <netinet/ip_var.h>
54 #include <netinet6/ip6_var.h>
55 #include <netinet/tcp.h>
56 #include <netinet/tcp_fsm.h>
57 #include <netinet/tcp_seq.h>
58 
59 #ifdef INET6
60 #include <netinet/ip6.h>
61 #endif /* INET6 */
62 
63 struct pf_frent {
64 	TAILQ_ENTRY(pf_frent)	fr_next;
65 	struct mbuf	*fe_m;
66 	uint16_t	fe_hdrlen;	/* ipv4 header length with ip options
67 					   ipv6, extension, fragment header */
68 	uint16_t	fe_extoff;	/* last extension header offset or 0 */
69 	uint16_t	fe_len;		/* fragment length */
70 	uint16_t	fe_off;		/* fragment offset */
71 	uint16_t	fe_mff;		/* more fragment flag */
72 };
73 
74 struct pf_fragment_cmp {
75 	struct pf_addr	frc_src;
76 	struct pf_addr	frc_dst;
77 	uint32_t	frc_id;
78 	sa_family_t	frc_af;
79 	uint8_t		frc_proto;
80 };
81 
82 struct pf_fragment {
83 	struct pf_fragment_cmp	fr_key;
84 #define fr_src	fr_key.frc_src
85 #define fr_dst	fr_key.frc_dst
86 #define fr_id	fr_key.frc_id
87 #define fr_af	fr_key.frc_af
88 #define fr_proto	fr_key.frc_proto
89 
90 	RB_ENTRY(pf_fragment) fr_entry;
91 	TAILQ_ENTRY(pf_fragment) frag_next;
92 	uint32_t	fr_timeout;
93 	uint16_t	fr_maxlen;	/* maximum length of single fragment */
94 	TAILQ_HEAD(pf_fragq, pf_frent) fr_queue;
95 };
96 
97 struct pf_fragment_tag {
98 	uint16_t	ft_hdrlen;	/* header length of reassembled pkt */
99 	uint16_t	ft_extoff;	/* last extension header offset or 0 */
100 	uint16_t	ft_maxlen;	/* maximum fragment payload length */
101 	uint32_t	ft_id;		/* fragment id */
102 };
103 
104 static struct mtx pf_frag_mtx;
105 MTX_SYSINIT(pf_frag_mtx, &pf_frag_mtx, "pf fragments", MTX_DEF);
106 #define PF_FRAG_LOCK()		mtx_lock(&pf_frag_mtx)
107 #define PF_FRAG_UNLOCK()	mtx_unlock(&pf_frag_mtx)
108 #define PF_FRAG_ASSERT()	mtx_assert(&pf_frag_mtx, MA_OWNED)
109 
110 VNET_DEFINE(uma_zone_t, pf_state_scrub_z);	/* XXX: shared with pfsync */
111 
112 static VNET_DEFINE(uma_zone_t, pf_frent_z);
113 #define	V_pf_frent_z	VNET(pf_frent_z)
114 static VNET_DEFINE(uma_zone_t, pf_frag_z);
115 #define	V_pf_frag_z	VNET(pf_frag_z)
116 
117 TAILQ_HEAD(pf_fragqueue, pf_fragment);
118 TAILQ_HEAD(pf_cachequeue, pf_fragment);
119 static VNET_DEFINE(struct pf_fragqueue,	pf_fragqueue);
120 #define	V_pf_fragqueue			VNET(pf_fragqueue)
121 RB_HEAD(pf_frag_tree, pf_fragment);
122 static VNET_DEFINE(struct pf_frag_tree,	pf_frag_tree);
123 #define	V_pf_frag_tree			VNET(pf_frag_tree)
124 static int		 pf_frag_compare(struct pf_fragment *,
125 			    struct pf_fragment *);
126 static RB_PROTOTYPE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
127 static RB_GENERATE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare);
128 
129 static void	pf_flush_fragments(void);
130 static void	pf_free_fragment(struct pf_fragment *);
131 static void	pf_remove_fragment(struct pf_fragment *);
132 static int	pf_normalize_tcpopt(struct pf_rule *, struct mbuf *,
133 		    struct tcphdr *, int, sa_family_t);
134 static struct pf_frent *pf_create_fragment(u_short *);
135 static struct pf_fragment *pf_find_fragment(struct pf_fragment_cmp *key,
136 		    struct pf_frag_tree *tree);
137 static struct pf_fragment *pf_fillup_fragment(struct pf_fragment_cmp *,
138 		    struct pf_frent *, u_short *);
139 static int	pf_isfull_fragment(struct pf_fragment *);
140 static struct mbuf *pf_join_fragment(struct pf_fragment *);
141 #ifdef INET
142 static void	pf_scrub_ip(struct mbuf **, uint32_t, uint8_t, uint8_t);
143 static int	pf_reassemble(struct mbuf **, struct ip *, int, u_short *);
144 #endif	/* INET */
145 #ifdef INET6
146 static int	pf_reassemble6(struct mbuf **, struct ip6_hdr *,
147 		    struct ip6_frag *, uint16_t, uint16_t, u_short *);
148 static void	pf_scrub_ip6(struct mbuf **, uint8_t);
149 #endif	/* INET6 */
150 
151 #define	DPFPRINTF(x) do {				\
152 	if (V_pf_status.debug >= PF_DEBUG_MISC) {	\
153 		printf("%s: ", __func__);		\
154 		printf x ;				\
155 	}						\
156 } while(0)
157 
158 #ifdef INET
159 static void
160 pf_ip2key(struct ip *ip, int dir, struct pf_fragment_cmp *key)
161 {
162 
163 	key->frc_src.v4 = ip->ip_src;
164 	key->frc_dst.v4 = ip->ip_dst;
165 	key->frc_af = AF_INET;
166 	key->frc_proto = ip->ip_p;
167 	key->frc_id = ip->ip_id;
168 }
169 #endif	/* INET */
170 
171 void
172 pf_normalize_init(void)
173 {
174 
175 	V_pf_frag_z = uma_zcreate("pf frags", sizeof(struct pf_fragment),
176 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
177 	V_pf_frent_z = uma_zcreate("pf frag entries", sizeof(struct pf_frent),
178 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
179 	V_pf_state_scrub_z = uma_zcreate("pf state scrubs",
180 	    sizeof(struct pf_state_scrub),  NULL, NULL, NULL, NULL,
181 	    UMA_ALIGN_PTR, 0);
182 
183 	V_pf_limits[PF_LIMIT_FRAGS].zone = V_pf_frent_z;
184 	V_pf_limits[PF_LIMIT_FRAGS].limit = PFFRAG_FRENT_HIWAT;
185 	uma_zone_set_max(V_pf_frent_z, PFFRAG_FRENT_HIWAT);
186 	uma_zone_set_warning(V_pf_frent_z, "PF frag entries limit reached");
187 
188 	TAILQ_INIT(&V_pf_fragqueue);
189 }
190 
191 void
192 pf_normalize_cleanup(void)
193 {
194 
195 	uma_zdestroy(V_pf_state_scrub_z);
196 	uma_zdestroy(V_pf_frent_z);
197 	uma_zdestroy(V_pf_frag_z);
198 }
199 
200 static int
201 pf_frag_compare(struct pf_fragment *a, struct pf_fragment *b)
202 {
203 	int	diff;
204 
205 	if ((diff = a->fr_id - b->fr_id) != 0)
206 		return (diff);
207 	if ((diff = a->fr_proto - b->fr_proto) != 0)
208 		return (diff);
209 	if ((diff = a->fr_af - b->fr_af) != 0)
210 		return (diff);
211 	if ((diff = pf_addr_cmp(&a->fr_src, &b->fr_src, a->fr_af)) != 0)
212 		return (diff);
213 	if ((diff = pf_addr_cmp(&a->fr_dst, &b->fr_dst, a->fr_af)) != 0)
214 		return (diff);
215 	return (0);
216 }
217 
218 void
219 pf_purge_expired_fragments(void)
220 {
221 	u_int32_t	expire = time_uptime -
222 			    V_pf_default_rule.timeout[PFTM_FRAG];
223 
224 	pf_purge_fragments(expire);
225 }
226 
227 void
228 pf_purge_fragments(uint32_t expire)
229 {
230 	struct pf_fragment	*frag;
231 
232 	PF_FRAG_LOCK();
233 	while ((frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue)) != NULL) {
234 		if (frag->fr_timeout > expire)
235 			break;
236 
237 		DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag));
238 		pf_free_fragment(frag);
239 	}
240 
241 	PF_FRAG_UNLOCK();
242 }
243 
244 /*
245  * Try to flush old fragments to make space for new ones
246  */
247 static void
248 pf_flush_fragments(void)
249 {
250 	struct pf_fragment	*frag;
251 	int			 goal;
252 
253 	PF_FRAG_ASSERT();
254 
255 	goal = uma_zone_get_cur(V_pf_frent_z) * 9 / 10;
256 	DPFPRINTF(("trying to free %d frag entriess\n", goal));
257 	while (goal < uma_zone_get_cur(V_pf_frent_z)) {
258 		frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue);
259 		if (frag)
260 			pf_free_fragment(frag);
261 		else
262 			break;
263 	}
264 }
265 
266 /* Frees the fragments and all associated entries */
267 static void
268 pf_free_fragment(struct pf_fragment *frag)
269 {
270 	struct pf_frent		*frent;
271 
272 	PF_FRAG_ASSERT();
273 
274 	/* Free all fragments */
275 	for (frent = TAILQ_FIRST(&frag->fr_queue); frent;
276 	    frent = TAILQ_FIRST(&frag->fr_queue)) {
277 		TAILQ_REMOVE(&frag->fr_queue, frent, fr_next);
278 
279 		m_freem(frent->fe_m);
280 		uma_zfree(V_pf_frent_z, frent);
281 	}
282 
283 	pf_remove_fragment(frag);
284 }
285 
286 static struct pf_fragment *
287 pf_find_fragment(struct pf_fragment_cmp *key, struct pf_frag_tree *tree)
288 {
289 	struct pf_fragment	*frag;
290 
291 	PF_FRAG_ASSERT();
292 
293 	frag = RB_FIND(pf_frag_tree, tree, (struct pf_fragment *)key);
294 	if (frag != NULL) {
295 		/* XXX Are we sure we want to update the timeout? */
296 		frag->fr_timeout = time_uptime;
297 		TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next);
298 		TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next);
299 	}
300 
301 	return (frag);
302 }
303 
304 /* Removes a fragment from the fragment queue and frees the fragment */
305 static void
306 pf_remove_fragment(struct pf_fragment *frag)
307 {
308 
309 	PF_FRAG_ASSERT();
310 
311 	RB_REMOVE(pf_frag_tree, &V_pf_frag_tree, frag);
312 	TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next);
313 	uma_zfree(V_pf_frag_z, frag);
314 }
315 
316 static struct pf_frent *
317 pf_create_fragment(u_short *reason)
318 {
319 	struct pf_frent *frent;
320 
321 	PF_FRAG_ASSERT();
322 
323 	frent = uma_zalloc(V_pf_frent_z, M_NOWAIT);
324 	if (frent == NULL) {
325 		pf_flush_fragments();
326 		frent = uma_zalloc(V_pf_frent_z, M_NOWAIT);
327 		if (frent == NULL) {
328 			REASON_SET(reason, PFRES_MEMORY);
329 			return (NULL);
330 		}
331 	}
332 
333 	return (frent);
334 }
335 
336 static struct pf_fragment *
337 pf_fillup_fragment(struct pf_fragment_cmp *key, struct pf_frent *frent,
338 		u_short *reason)
339 {
340 	struct pf_frent		*after, *next, *prev;
341 	struct pf_fragment	*frag;
342 	uint16_t		total;
343 
344 	PF_FRAG_ASSERT();
345 
346 	/* No empty fragments. */
347 	if (frent->fe_len == 0) {
348 		DPFPRINTF(("bad fragment: len 0"));
349 		goto bad_fragment;
350 	}
351 
352 	/* All fragments are 8 byte aligned. */
353 	if (frent->fe_mff && (frent->fe_len & 0x7)) {
354 		DPFPRINTF(("bad fragment: mff and len %d", frent->fe_len));
355 		goto bad_fragment;
356 	}
357 
358 	/* Respect maximum length, IP_MAXPACKET == IPV6_MAXPACKET. */
359 	if (frent->fe_off + frent->fe_len > IP_MAXPACKET) {
360 		DPFPRINTF(("bad fragment: max packet %d",
361 		    frent->fe_off + frent->fe_len));
362 		goto bad_fragment;
363 	}
364 
365 	DPFPRINTF((key->frc_af == AF_INET ?
366 	    "reass frag %d @ %d-%d" : "reass frag %#08x @ %d-%d",
367 	    key->frc_id, frent->fe_off, frent->fe_off + frent->fe_len));
368 
369 	/* Fully buffer all of the fragments in this fragment queue. */
370 	frag = pf_find_fragment(key, &V_pf_frag_tree);
371 
372 	/* Create a new reassembly queue for this packet. */
373 	if (frag == NULL) {
374 		frag = uma_zalloc(V_pf_frag_z, M_NOWAIT);
375 		if (frag == NULL) {
376 			pf_flush_fragments();
377 			frag = uma_zalloc(V_pf_frag_z, M_NOWAIT);
378 			if (frag == NULL) {
379 				REASON_SET(reason, PFRES_MEMORY);
380 				goto drop_fragment;
381 			}
382 		}
383 
384 		*(struct pf_fragment_cmp *)frag = *key;
385 		frag->fr_timeout = time_uptime;
386 		frag->fr_maxlen = frent->fe_len;
387 		TAILQ_INIT(&frag->fr_queue);
388 
389 		RB_INSERT(pf_frag_tree, &V_pf_frag_tree, frag);
390 		TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next);
391 
392 		/* We do not have a previous fragment. */
393 		TAILQ_INSERT_HEAD(&frag->fr_queue, frent, fr_next);
394 
395 		return (frag);
396 	}
397 
398 	KASSERT(!TAILQ_EMPTY(&frag->fr_queue), ("!TAILQ_EMPTY()->fr_queue"));
399 
400 	/* Remember maximum fragment len for refragmentation. */
401 	if (frent->fe_len > frag->fr_maxlen)
402 		frag->fr_maxlen = frent->fe_len;
403 
404 	/* Maximum data we have seen already. */
405 	total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
406 		TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
407 
408 	/* Non terminal fragments must have more fragments flag. */
409 	if (frent->fe_off + frent->fe_len < total && !frent->fe_mff)
410 		goto bad_fragment;
411 
412 	/* Check if we saw the last fragment already. */
413 	if (!TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_mff) {
414 		if (frent->fe_off + frent->fe_len > total ||
415 		    (frent->fe_off + frent->fe_len == total && frent->fe_mff))
416 			goto bad_fragment;
417 	} else {
418 		if (frent->fe_off + frent->fe_len == total && !frent->fe_mff)
419 			goto bad_fragment;
420 	}
421 
422 	/* Find a fragment after the current one. */
423 	prev = NULL;
424 	TAILQ_FOREACH(after, &frag->fr_queue, fr_next) {
425 		if (after->fe_off > frent->fe_off)
426 			break;
427 		prev = after;
428 	}
429 
430 	KASSERT(prev != NULL || after != NULL,
431 	    ("prev != NULL || after != NULL"));
432 
433 	if (prev != NULL && prev->fe_off + prev->fe_len > frent->fe_off) {
434 		uint16_t precut;
435 
436 		precut = prev->fe_off + prev->fe_len - frent->fe_off;
437 		if (precut >= frent->fe_len)
438 			goto bad_fragment;
439 		DPFPRINTF(("overlap -%d", precut));
440 		m_adj(frent->fe_m, precut);
441 		frent->fe_off += precut;
442 		frent->fe_len -= precut;
443 	}
444 
445 	for (; after != NULL && frent->fe_off + frent->fe_len > after->fe_off;
446 	    after = next) {
447 		uint16_t aftercut;
448 
449 		aftercut = frent->fe_off + frent->fe_len - after->fe_off;
450 		DPFPRINTF(("adjust overlap %d", aftercut));
451 		if (aftercut < after->fe_len) {
452 			m_adj(after->fe_m, aftercut);
453 			after->fe_off += aftercut;
454 			after->fe_len -= aftercut;
455 			break;
456 		}
457 
458 		/* This fragment is completely overlapped, lose it. */
459 		next = TAILQ_NEXT(after, fr_next);
460 		m_freem(after->fe_m);
461 		TAILQ_REMOVE(&frag->fr_queue, after, fr_next);
462 		uma_zfree(V_pf_frent_z, after);
463 	}
464 
465 	if (prev == NULL)
466 		TAILQ_INSERT_HEAD(&frag->fr_queue, frent, fr_next);
467 	else
468 		TAILQ_INSERT_AFTER(&frag->fr_queue, prev, frent, fr_next);
469 
470 	return (frag);
471 
472 bad_fragment:
473 	REASON_SET(reason, PFRES_FRAG);
474 drop_fragment:
475 	uma_zfree(V_pf_frent_z, frent);
476 	return (NULL);
477 }
478 
479 static int
480 pf_isfull_fragment(struct pf_fragment *frag)
481 {
482 	struct pf_frent	*frent, *next;
483 	uint16_t off, total;
484 
485 	/* Check if we are completely reassembled */
486 	if (TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_mff)
487 		return (0);
488 
489 	/* Maximum data we have seen already */
490 	total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
491 		TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
492 
493 	/* Check if we have all the data */
494 	off = 0;
495 	for (frent = TAILQ_FIRST(&frag->fr_queue); frent; frent = next) {
496 		next = TAILQ_NEXT(frent, fr_next);
497 
498 		off += frent->fe_len;
499 		if (off < total && (next == NULL || next->fe_off != off)) {
500 			DPFPRINTF(("missing fragment at %d, next %d, total %d",
501 			    off, next == NULL ? -1 : next->fe_off, total));
502 			return (0);
503 		}
504 	}
505 	DPFPRINTF(("%d < %d?", off, total));
506 	if (off < total)
507 		return (0);
508 	KASSERT(off == total, ("off == total"));
509 
510 	return (1);
511 }
512 
513 static struct mbuf *
514 pf_join_fragment(struct pf_fragment *frag)
515 {
516 	struct mbuf *m, *m2;
517 	struct pf_frent	*frent, *next;
518 
519 	frent = TAILQ_FIRST(&frag->fr_queue);
520 	next = TAILQ_NEXT(frent, fr_next);
521 
522 	m = frent->fe_m;
523 	m_adj(m, (frent->fe_hdrlen + frent->fe_len) - m->m_pkthdr.len);
524 	uma_zfree(V_pf_frent_z, frent);
525 	for (frent = next; frent != NULL; frent = next) {
526 		next = TAILQ_NEXT(frent, fr_next);
527 
528 		m2 = frent->fe_m;
529 		/* Strip off ip header. */
530 		m_adj(m2, frent->fe_hdrlen);
531 		/* Strip off any trailing bytes. */
532 		m_adj(m2, frent->fe_len - m2->m_pkthdr.len);
533 
534 		uma_zfree(V_pf_frent_z, frent);
535 		m_cat(m, m2);
536 	}
537 
538 	/* Remove from fragment queue. */
539 	pf_remove_fragment(frag);
540 
541 	return (m);
542 }
543 
544 #ifdef INET
545 static int
546 pf_reassemble(struct mbuf **m0, struct ip *ip, int dir, u_short *reason)
547 {
548 	struct mbuf		*m = *m0;
549 	struct pf_frent		*frent;
550 	struct pf_fragment	*frag;
551 	struct pf_fragment_cmp	key;
552 	uint16_t		total, hdrlen;
553 
554 	/* Get an entry for the fragment queue */
555 	if ((frent = pf_create_fragment(reason)) == NULL)
556 		return (PF_DROP);
557 
558 	frent->fe_m = m;
559 	frent->fe_hdrlen = ip->ip_hl << 2;
560 	frent->fe_extoff = 0;
561 	frent->fe_len = ntohs(ip->ip_len) - (ip->ip_hl << 2);
562 	frent->fe_off = (ntohs(ip->ip_off) & IP_OFFMASK) << 3;
563 	frent->fe_mff = ntohs(ip->ip_off) & IP_MF;
564 
565 	pf_ip2key(ip, dir, &key);
566 
567 	if ((frag = pf_fillup_fragment(&key, frent, reason)) == NULL)
568 		return (PF_DROP);
569 
570 	/* The mbuf is part of the fragment entry, no direct free or access */
571 	m = *m0 = NULL;
572 
573 	if (!pf_isfull_fragment(frag))
574 		return (PF_PASS);  /* drop because *m0 is NULL, no error */
575 
576 	/* We have all the data */
577 	frent = TAILQ_FIRST(&frag->fr_queue);
578 	KASSERT(frent != NULL, ("frent != NULL"));
579 	total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
580 		TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
581 	hdrlen = frent->fe_hdrlen;
582 
583 	m = *m0 = pf_join_fragment(frag);
584 	frag = NULL;
585 
586 	if (m->m_flags & M_PKTHDR) {
587 		int plen = 0;
588 		for (m = *m0; m; m = m->m_next)
589 			plen += m->m_len;
590 		m = *m0;
591 		m->m_pkthdr.len = plen;
592 	}
593 
594 	ip = mtod(m, struct ip *);
595 	ip->ip_len = htons(hdrlen + total);
596 	ip->ip_off &= ~(IP_MF|IP_OFFMASK);
597 
598 	if (hdrlen + total > IP_MAXPACKET) {
599 		DPFPRINTF(("drop: too big: %d", total));
600 		ip->ip_len = 0;
601 		REASON_SET(reason, PFRES_SHORT);
602 		/* PF_DROP requires a valid mbuf *m0 in pf_test() */
603 		return (PF_DROP);
604 	}
605 
606 	DPFPRINTF(("complete: %p(%d)\n", m, ntohs(ip->ip_len)));
607 	return (PF_PASS);
608 }
609 #endif	/* INET */
610 
611 #ifdef INET6
612 static int
613 pf_reassemble6(struct mbuf **m0, struct ip6_hdr *ip6, struct ip6_frag *fraghdr,
614     uint16_t hdrlen, uint16_t extoff, u_short *reason)
615 {
616 	struct mbuf		*m = *m0;
617 	struct pf_frent		*frent;
618 	struct pf_fragment	*frag;
619 	struct pf_fragment_cmp	 key;
620 	struct m_tag		*mtag;
621 	struct pf_fragment_tag	*ftag;
622 	int			 off;
623 	uint32_t		 frag_id;
624 	uint16_t		 total, maxlen;
625 	uint8_t			 proto;
626 
627 	PF_FRAG_LOCK();
628 
629 	/* Get an entry for the fragment queue. */
630 	if ((frent = pf_create_fragment(reason)) == NULL) {
631 		PF_FRAG_UNLOCK();
632 		return (PF_DROP);
633 	}
634 
635 	frent->fe_m = m;
636 	frent->fe_hdrlen = hdrlen;
637 	frent->fe_extoff = extoff;
638 	frent->fe_len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - hdrlen;
639 	frent->fe_off = ntohs(fraghdr->ip6f_offlg & IP6F_OFF_MASK);
640 	frent->fe_mff = fraghdr->ip6f_offlg & IP6F_MORE_FRAG;
641 
642 	key.frc_src.v6 = ip6->ip6_src;
643 	key.frc_dst.v6 = ip6->ip6_dst;
644 	key.frc_af = AF_INET6;
645 	/* Only the first fragment's protocol is relevant. */
646 	key.frc_proto = 0;
647 	key.frc_id = fraghdr->ip6f_ident;
648 
649 	if ((frag = pf_fillup_fragment(&key, frent, reason)) == NULL) {
650 		PF_FRAG_UNLOCK();
651 		return (PF_DROP);
652 	}
653 
654 	/* The mbuf is part of the fragment entry, no direct free or access. */
655 	m = *m0 = NULL;
656 
657 	if (!pf_isfull_fragment(frag)) {
658 		PF_FRAG_UNLOCK();
659 		return (PF_PASS);  /* Drop because *m0 is NULL, no error. */
660 	}
661 
662 	/* We have all the data. */
663 	extoff = frent->fe_extoff;
664 	maxlen = frag->fr_maxlen;
665 	frag_id = frag->fr_id;
666 	frent = TAILQ_FIRST(&frag->fr_queue);
667 	KASSERT(frent != NULL, ("frent != NULL"));
668 	total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off +
669 		TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len;
670 	hdrlen = frent->fe_hdrlen - sizeof(struct ip6_frag);
671 
672 	m = *m0 = pf_join_fragment(frag);
673 	frag = NULL;
674 
675 	PF_FRAG_UNLOCK();
676 
677 	/* Take protocol from first fragment header. */
678 	m = m_getptr(m, hdrlen + offsetof(struct ip6_frag, ip6f_nxt), &off);
679 	KASSERT(m, ("%s: short mbuf chain", __func__));
680 	proto = *(mtod(m, caddr_t) + off);
681 	m = *m0;
682 
683 	/* Delete frag6 header */
684 	if (ip6_deletefraghdr(m, hdrlen, M_NOWAIT) != 0)
685 		goto fail;
686 
687 	if (m->m_flags & M_PKTHDR) {
688 		int plen = 0;
689 		for (m = *m0; m; m = m->m_next)
690 			plen += m->m_len;
691 		m = *m0;
692 		m->m_pkthdr.len = plen;
693 	}
694 
695 	if ((mtag = m_tag_get(PF_REASSEMBLED, sizeof(struct pf_fragment_tag),
696 	    M_NOWAIT)) == NULL)
697 		goto fail;
698 	ftag = (struct pf_fragment_tag *)(mtag + 1);
699 	ftag->ft_hdrlen = hdrlen;
700 	ftag->ft_extoff = extoff;
701 	ftag->ft_maxlen = maxlen;
702 	ftag->ft_id = frag_id;
703 	m_tag_prepend(m, mtag);
704 
705 	ip6 = mtod(m, struct ip6_hdr *);
706 	ip6->ip6_plen = htons(hdrlen - sizeof(struct ip6_hdr) + total);
707 	if (extoff) {
708 		/* Write protocol into next field of last extension header. */
709 		m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt),
710 		    &off);
711 		KASSERT(m, ("%s: short mbuf chain", __func__));
712 		*(mtod(m, char *) + off) = proto;
713 		m = *m0;
714 	} else
715 		ip6->ip6_nxt = proto;
716 
717 	if (hdrlen - sizeof(struct ip6_hdr) + total > IPV6_MAXPACKET) {
718 		DPFPRINTF(("drop: too big: %d", total));
719 		ip6->ip6_plen = 0;
720 		REASON_SET(reason, PFRES_SHORT);
721 		/* PF_DROP requires a valid mbuf *m0 in pf_test6(). */
722 		return (PF_DROP);
723 	}
724 
725 	DPFPRINTF(("complete: %p(%d)", m, ntohs(ip6->ip6_plen)));
726 	return (PF_PASS);
727 
728 fail:
729 	REASON_SET(reason, PFRES_MEMORY);
730 	/* PF_DROP requires a valid mbuf *m0 in pf_test6(), will free later. */
731 	return (PF_DROP);
732 }
733 #endif	/* INET6 */
734 
735 #ifdef INET6
736 int
737 pf_refragment6(struct ifnet *ifp, struct mbuf **m0, struct m_tag *mtag)
738 {
739 	struct mbuf		*m = *m0, *t;
740 	struct pf_fragment_tag	*ftag = (struct pf_fragment_tag *)(mtag + 1);
741 	struct pf_pdesc		 pd;
742 	uint32_t		 frag_id;
743 	uint16_t		 hdrlen, extoff, maxlen;
744 	uint8_t			 proto;
745 	int			 error, action;
746 
747 	hdrlen = ftag->ft_hdrlen;
748 	extoff = ftag->ft_extoff;
749 	maxlen = ftag->ft_maxlen;
750 	frag_id = ftag->ft_id;
751 	m_tag_delete(m, mtag);
752 	mtag = NULL;
753 	ftag = NULL;
754 
755 	if (extoff) {
756 		int off;
757 
758 		/* Use protocol from next field of last extension header */
759 		m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt),
760 		    &off);
761 		KASSERT((m != NULL), ("pf_refragment6: short mbuf chain"));
762 		proto = *(mtod(m, caddr_t) + off);
763 		*(mtod(m, char *) + off) = IPPROTO_FRAGMENT;
764 		m = *m0;
765 	} else {
766 		struct ip6_hdr *hdr;
767 
768 		hdr = mtod(m, struct ip6_hdr *);
769 		proto = hdr->ip6_nxt;
770 		hdr->ip6_nxt = IPPROTO_FRAGMENT;
771 	}
772 
773 	/* The MTU must be a multiple of 8 bytes, or we risk doing the
774 	 * fragmentation wrong. */
775 	maxlen = maxlen & ~7;
776 
777 	/*
778 	 * Maxlen may be less than 8 if there was only a single
779 	 * fragment.  As it was fragmented before, add a fragment
780 	 * header also for a single fragment.  If total or maxlen
781 	 * is less than 8, ip6_fragment() will return EMSGSIZE and
782 	 * we drop the packet.
783 	 */
784 	error = ip6_fragment(ifp, m, hdrlen, proto, maxlen, frag_id);
785 	m = (*m0)->m_nextpkt;
786 	(*m0)->m_nextpkt = NULL;
787 	if (error == 0) {
788 		/* The first mbuf contains the unfragmented packet. */
789 		m_freem(*m0);
790 		*m0 = NULL;
791 		action = PF_PASS;
792 	} else {
793 		/* Drop expects an mbuf to free. */
794 		DPFPRINTF(("refragment error %d", error));
795 		action = PF_DROP;
796 	}
797 	for (t = m; m; m = t) {
798 		t = m->m_nextpkt;
799 		m->m_nextpkt = NULL;
800 		m->m_flags |= M_SKIP_FIREWALL;
801 		memset(&pd, 0, sizeof(pd));
802 		pd.pf_mtag = pf_find_mtag(m);
803 		if (error == 0)
804 			ip6_forward(m, 0);
805 		else
806 			m_freem(m);
807 	}
808 
809 	return (action);
810 }
811 #endif /* INET6 */
812 
813 #ifdef INET
814 int
815 pf_normalize_ip(struct mbuf **m0, int dir, struct pfi_kif *kif, u_short *reason,
816     struct pf_pdesc *pd)
817 {
818 	struct mbuf		*m = *m0;
819 	struct pf_rule		*r;
820 	struct ip		*h = mtod(m, struct ip *);
821 	int			 mff = (ntohs(h->ip_off) & IP_MF);
822 	int			 hlen = h->ip_hl << 2;
823 	u_int16_t		 fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
824 	u_int16_t		 max;
825 	int			 ip_len;
826 	int			 ip_off;
827 	int			 tag = -1;
828 	int			 verdict;
829 
830 	PF_RULES_RASSERT();
831 
832 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
833 	while (r != NULL) {
834 		r->evaluations++;
835 		if (pfi_kif_match(r->kif, kif) == r->ifnot)
836 			r = r->skip[PF_SKIP_IFP].ptr;
837 		else if (r->direction && r->direction != dir)
838 			r = r->skip[PF_SKIP_DIR].ptr;
839 		else if (r->af && r->af != AF_INET)
840 			r = r->skip[PF_SKIP_AF].ptr;
841 		else if (r->proto && r->proto != h->ip_p)
842 			r = r->skip[PF_SKIP_PROTO].ptr;
843 		else if (PF_MISMATCHAW(&r->src.addr,
844 		    (struct pf_addr *)&h->ip_src.s_addr, AF_INET,
845 		    r->src.neg, kif, M_GETFIB(m)))
846 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
847 		else if (PF_MISMATCHAW(&r->dst.addr,
848 		    (struct pf_addr *)&h->ip_dst.s_addr, AF_INET,
849 		    r->dst.neg, NULL, M_GETFIB(m)))
850 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
851 		else if (r->match_tag && !pf_match_tag(m, r, &tag,
852 		    pd->pf_mtag ? pd->pf_mtag->tag : 0))
853 			r = TAILQ_NEXT(r, entries);
854 		else
855 			break;
856 	}
857 
858 	if (r == NULL || r->action == PF_NOSCRUB)
859 		return (PF_PASS);
860 	else {
861 		r->packets[dir == PF_OUT]++;
862 		r->bytes[dir == PF_OUT] += pd->tot_len;
863 	}
864 
865 	/* Check for illegal packets */
866 	if (hlen < (int)sizeof(struct ip)) {
867 		REASON_SET(reason, PFRES_NORM);
868 		goto drop;
869 	}
870 
871 	if (hlen > ntohs(h->ip_len)) {
872 		REASON_SET(reason, PFRES_NORM);
873 		goto drop;
874 	}
875 
876 	/* Clear IP_DF if the rule uses the no-df option */
877 	if (r->rule_flag & PFRULE_NODF && h->ip_off & htons(IP_DF)) {
878 		u_int16_t ip_off = h->ip_off;
879 
880 		h->ip_off &= htons(~IP_DF);
881 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
882 	}
883 
884 	/* We will need other tests here */
885 	if (!fragoff && !mff)
886 		goto no_fragment;
887 
888 	/* We're dealing with a fragment now. Don't allow fragments
889 	 * with IP_DF to enter the cache. If the flag was cleared by
890 	 * no-df above, fine. Otherwise drop it.
891 	 */
892 	if (h->ip_off & htons(IP_DF)) {
893 		DPFPRINTF(("IP_DF\n"));
894 		goto bad;
895 	}
896 
897 	ip_len = ntohs(h->ip_len) - hlen;
898 	ip_off = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
899 
900 	/* All fragments are 8 byte aligned */
901 	if (mff && (ip_len & 0x7)) {
902 		DPFPRINTF(("mff and %d\n", ip_len));
903 		goto bad;
904 	}
905 
906 	/* Respect maximum length */
907 	if (fragoff + ip_len > IP_MAXPACKET) {
908 		DPFPRINTF(("max packet %d\n", fragoff + ip_len));
909 		goto bad;
910 	}
911 	max = fragoff + ip_len;
912 
913 	/* Fully buffer all of the fragments
914 	 * Might return a completely reassembled mbuf, or NULL */
915 	PF_FRAG_LOCK();
916 	DPFPRINTF(("reass frag %d @ %d-%d\n", h->ip_id, fragoff, max));
917 	verdict = pf_reassemble(m0, h, dir, reason);
918 	PF_FRAG_UNLOCK();
919 
920 	if (verdict != PF_PASS)
921 		return (PF_DROP);
922 
923 	m = *m0;
924 	if (m == NULL)
925 		return (PF_DROP);
926 
927 	h = mtod(m, struct ip *);
928 
929  no_fragment:
930 	/* At this point, only IP_DF is allowed in ip_off */
931 	if (h->ip_off & ~htons(IP_DF)) {
932 		u_int16_t ip_off = h->ip_off;
933 
934 		h->ip_off &= htons(IP_DF);
935 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
936 	}
937 
938 	pf_scrub_ip(&m, r->rule_flag, r->min_ttl, r->set_tos);
939 
940 	return (PF_PASS);
941 
942  bad:
943 	DPFPRINTF(("dropping bad fragment\n"));
944 	REASON_SET(reason, PFRES_FRAG);
945  drop:
946 	if (r != NULL && r->log)
947 		PFLOG_PACKET(kif, m, AF_INET, dir, *reason, r, NULL, NULL, pd,
948 		    1);
949 
950 	return (PF_DROP);
951 }
952 #endif
953 
954 #ifdef INET6
955 int
956 pf_normalize_ip6(struct mbuf **m0, int dir, struct pfi_kif *kif,
957     u_short *reason, struct pf_pdesc *pd)
958 {
959 	struct mbuf		*m = *m0;
960 	struct pf_rule		*r;
961 	struct ip6_hdr		*h = mtod(m, struct ip6_hdr *);
962 	int			 extoff;
963 	int			 off;
964 	struct ip6_ext		 ext;
965 	struct ip6_opt		 opt;
966 	struct ip6_opt_jumbo	 jumbo;
967 	struct ip6_frag		 frag;
968 	u_int32_t		 jumbolen = 0, plen;
969 	int			 optend;
970 	int			 ooff;
971 	u_int8_t		 proto;
972 	int			 terminal;
973 
974 	PF_RULES_RASSERT();
975 
976 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
977 	while (r != NULL) {
978 		r->evaluations++;
979 		if (pfi_kif_match(r->kif, kif) == r->ifnot)
980 			r = r->skip[PF_SKIP_IFP].ptr;
981 		else if (r->direction && r->direction != dir)
982 			r = r->skip[PF_SKIP_DIR].ptr;
983 		else if (r->af && r->af != AF_INET6)
984 			r = r->skip[PF_SKIP_AF].ptr;
985 #if 0 /* header chain! */
986 		else if (r->proto && r->proto != h->ip6_nxt)
987 			r = r->skip[PF_SKIP_PROTO].ptr;
988 #endif
989 		else if (PF_MISMATCHAW(&r->src.addr,
990 		    (struct pf_addr *)&h->ip6_src, AF_INET6,
991 		    r->src.neg, kif, M_GETFIB(m)))
992 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
993 		else if (PF_MISMATCHAW(&r->dst.addr,
994 		    (struct pf_addr *)&h->ip6_dst, AF_INET6,
995 		    r->dst.neg, NULL, M_GETFIB(m)))
996 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
997 		else
998 			break;
999 	}
1000 
1001 	if (r == NULL || r->action == PF_NOSCRUB)
1002 		return (PF_PASS);
1003 	else {
1004 		r->packets[dir == PF_OUT]++;
1005 		r->bytes[dir == PF_OUT] += pd->tot_len;
1006 	}
1007 
1008 	/* Check for illegal packets */
1009 	if (sizeof(struct ip6_hdr) + IPV6_MAXPACKET < m->m_pkthdr.len)
1010 		goto drop;
1011 
1012 	extoff = 0;
1013 	off = sizeof(struct ip6_hdr);
1014 	proto = h->ip6_nxt;
1015 	terminal = 0;
1016 	do {
1017 		switch (proto) {
1018 		case IPPROTO_FRAGMENT:
1019 			goto fragment;
1020 			break;
1021 		case IPPROTO_AH:
1022 		case IPPROTO_ROUTING:
1023 		case IPPROTO_DSTOPTS:
1024 			if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL,
1025 			    NULL, AF_INET6))
1026 				goto shortpkt;
1027 			extoff = off;
1028 			if (proto == IPPROTO_AH)
1029 				off += (ext.ip6e_len + 2) * 4;
1030 			else
1031 				off += (ext.ip6e_len + 1) * 8;
1032 			proto = ext.ip6e_nxt;
1033 			break;
1034 		case IPPROTO_HOPOPTS:
1035 			if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL,
1036 			    NULL, AF_INET6))
1037 				goto shortpkt;
1038 			extoff = off;
1039 			optend = off + (ext.ip6e_len + 1) * 8;
1040 			ooff = off + sizeof(ext);
1041 			do {
1042 				if (!pf_pull_hdr(m, ooff, &opt.ip6o_type,
1043 				    sizeof(opt.ip6o_type), NULL, NULL,
1044 				    AF_INET6))
1045 					goto shortpkt;
1046 				if (opt.ip6o_type == IP6OPT_PAD1) {
1047 					ooff++;
1048 					continue;
1049 				}
1050 				if (!pf_pull_hdr(m, ooff, &opt, sizeof(opt),
1051 				    NULL, NULL, AF_INET6))
1052 					goto shortpkt;
1053 				if (ooff + sizeof(opt) + opt.ip6o_len > optend)
1054 					goto drop;
1055 				switch (opt.ip6o_type) {
1056 				case IP6OPT_JUMBO:
1057 					if (h->ip6_plen != 0)
1058 						goto drop;
1059 					if (!pf_pull_hdr(m, ooff, &jumbo,
1060 					    sizeof(jumbo), NULL, NULL,
1061 					    AF_INET6))
1062 						goto shortpkt;
1063 					memcpy(&jumbolen, jumbo.ip6oj_jumbo_len,
1064 					    sizeof(jumbolen));
1065 					jumbolen = ntohl(jumbolen);
1066 					if (jumbolen <= IPV6_MAXPACKET)
1067 						goto drop;
1068 					if (sizeof(struct ip6_hdr) + jumbolen !=
1069 					    m->m_pkthdr.len)
1070 						goto drop;
1071 					break;
1072 				default:
1073 					break;
1074 				}
1075 				ooff += sizeof(opt) + opt.ip6o_len;
1076 			} while (ooff < optend);
1077 
1078 			off = optend;
1079 			proto = ext.ip6e_nxt;
1080 			break;
1081 		default:
1082 			terminal = 1;
1083 			break;
1084 		}
1085 	} while (!terminal);
1086 
1087 	/* jumbo payload option must be present, or plen > 0 */
1088 	if (ntohs(h->ip6_plen) == 0)
1089 		plen = jumbolen;
1090 	else
1091 		plen = ntohs(h->ip6_plen);
1092 	if (plen == 0)
1093 		goto drop;
1094 	if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len)
1095 		goto shortpkt;
1096 
1097 	pf_scrub_ip6(&m, r->min_ttl);
1098 
1099 	return (PF_PASS);
1100 
1101  fragment:
1102 	/* Jumbo payload packets cannot be fragmented. */
1103 	plen = ntohs(h->ip6_plen);
1104 	if (plen == 0 || jumbolen)
1105 		goto drop;
1106 	if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len)
1107 		goto shortpkt;
1108 
1109 	if (!pf_pull_hdr(m, off, &frag, sizeof(frag), NULL, NULL, AF_INET6))
1110 		goto shortpkt;
1111 
1112 	/* Offset now points to data portion. */
1113 	off += sizeof(frag);
1114 
1115 	/* Returns PF_DROP or *m0 is NULL or completely reassembled mbuf. */
1116 	if (pf_reassemble6(m0, h, &frag, off, extoff, reason) != PF_PASS)
1117 		return (PF_DROP);
1118 	m = *m0;
1119 	if (m == NULL)
1120 		return (PF_DROP);
1121 
1122 	pd->flags |= PFDESC_IP_REAS;
1123 	return (PF_PASS);
1124 
1125  shortpkt:
1126 	REASON_SET(reason, PFRES_SHORT);
1127 	if (r != NULL && r->log)
1128 		PFLOG_PACKET(kif, m, AF_INET6, dir, *reason, r, NULL, NULL, pd,
1129 		    1);
1130 	return (PF_DROP);
1131 
1132  drop:
1133 	REASON_SET(reason, PFRES_NORM);
1134 	if (r != NULL && r->log)
1135 		PFLOG_PACKET(kif, m, AF_INET6, dir, *reason, r, NULL, NULL, pd,
1136 		    1);
1137 	return (PF_DROP);
1138 }
1139 #endif /* INET6 */
1140 
1141 int
1142 pf_normalize_tcp(int dir, struct pfi_kif *kif, struct mbuf *m, int ipoff,
1143     int off, void *h, struct pf_pdesc *pd)
1144 {
1145 	struct pf_rule	*r, *rm = NULL;
1146 	struct tcphdr	*th = pd->hdr.tcp;
1147 	int		 rewrite = 0;
1148 	u_short		 reason;
1149 	u_int8_t	 flags;
1150 	sa_family_t	 af = pd->af;
1151 
1152 	PF_RULES_RASSERT();
1153 
1154 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr);
1155 	while (r != NULL) {
1156 		r->evaluations++;
1157 		if (pfi_kif_match(r->kif, kif) == r->ifnot)
1158 			r = r->skip[PF_SKIP_IFP].ptr;
1159 		else if (r->direction && r->direction != dir)
1160 			r = r->skip[PF_SKIP_DIR].ptr;
1161 		else if (r->af && r->af != af)
1162 			r = r->skip[PF_SKIP_AF].ptr;
1163 		else if (r->proto && r->proto != pd->proto)
1164 			r = r->skip[PF_SKIP_PROTO].ptr;
1165 		else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
1166 		    r->src.neg, kif, M_GETFIB(m)))
1167 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
1168 		else if (r->src.port_op && !pf_match_port(r->src.port_op,
1169 			    r->src.port[0], r->src.port[1], th->th_sport))
1170 			r = r->skip[PF_SKIP_SRC_PORT].ptr;
1171 		else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
1172 		    r->dst.neg, NULL, M_GETFIB(m)))
1173 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
1174 		else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
1175 			    r->dst.port[0], r->dst.port[1], th->th_dport))
1176 			r = r->skip[PF_SKIP_DST_PORT].ptr;
1177 		else if (r->os_fingerprint != PF_OSFP_ANY && !pf_osfp_match(
1178 			    pf_osfp_fingerprint(pd, m, off, th),
1179 			    r->os_fingerprint))
1180 			r = TAILQ_NEXT(r, entries);
1181 		else {
1182 			rm = r;
1183 			break;
1184 		}
1185 	}
1186 
1187 	if (rm == NULL || rm->action == PF_NOSCRUB)
1188 		return (PF_PASS);
1189 	else {
1190 		r->packets[dir == PF_OUT]++;
1191 		r->bytes[dir == PF_OUT] += pd->tot_len;
1192 	}
1193 
1194 	if (rm->rule_flag & PFRULE_REASSEMBLE_TCP)
1195 		pd->flags |= PFDESC_TCP_NORM;
1196 
1197 	flags = th->th_flags;
1198 	if (flags & TH_SYN) {
1199 		/* Illegal packet */
1200 		if (flags & TH_RST)
1201 			goto tcp_drop;
1202 
1203 		if (flags & TH_FIN)
1204 			goto tcp_drop;
1205 	} else {
1206 		/* Illegal packet */
1207 		if (!(flags & (TH_ACK|TH_RST)))
1208 			goto tcp_drop;
1209 	}
1210 
1211 	if (!(flags & TH_ACK)) {
1212 		/* These flags are only valid if ACK is set */
1213 		if ((flags & TH_FIN) || (flags & TH_PUSH) || (flags & TH_URG))
1214 			goto tcp_drop;
1215 	}
1216 
1217 	/* Check for illegal header length */
1218 	if (th->th_off < (sizeof(struct tcphdr) >> 2))
1219 		goto tcp_drop;
1220 
1221 	/* If flags changed, or reserved data set, then adjust */
1222 	if (flags != th->th_flags || th->th_x2 != 0) {
1223 		u_int16_t	ov, nv;
1224 
1225 		ov = *(u_int16_t *)(&th->th_ack + 1);
1226 		th->th_flags = flags;
1227 		th->th_x2 = 0;
1228 		nv = *(u_int16_t *)(&th->th_ack + 1);
1229 
1230 		th->th_sum = pf_proto_cksum_fixup(m, th->th_sum, ov, nv, 0);
1231 		rewrite = 1;
1232 	}
1233 
1234 	/* Remove urgent pointer, if TH_URG is not set */
1235 	if (!(flags & TH_URG) && th->th_urp) {
1236 		th->th_sum = pf_proto_cksum_fixup(m, th->th_sum, th->th_urp,
1237 		    0, 0);
1238 		th->th_urp = 0;
1239 		rewrite = 1;
1240 	}
1241 
1242 	/* Process options */
1243 	if (r->max_mss && pf_normalize_tcpopt(r, m, th, off, pd->af))
1244 		rewrite = 1;
1245 
1246 	/* copy back packet headers if we sanitized */
1247 	if (rewrite)
1248 		m_copyback(m, off, sizeof(*th), (caddr_t)th);
1249 
1250 	return (PF_PASS);
1251 
1252  tcp_drop:
1253 	REASON_SET(&reason, PFRES_NORM);
1254 	if (rm != NULL && r->log)
1255 		PFLOG_PACKET(kif, m, AF_INET, dir, reason, r, NULL, NULL, pd,
1256 		    1);
1257 	return (PF_DROP);
1258 }
1259 
1260 int
1261 pf_normalize_tcp_init(struct mbuf *m, int off, struct pf_pdesc *pd,
1262     struct tcphdr *th, struct pf_state_peer *src, struct pf_state_peer *dst)
1263 {
1264 	u_int32_t tsval, tsecr;
1265 	u_int8_t hdr[60];
1266 	u_int8_t *opt;
1267 
1268 	KASSERT((src->scrub == NULL),
1269 	    ("pf_normalize_tcp_init: src->scrub != NULL"));
1270 
1271 	src->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT);
1272 	if (src->scrub == NULL)
1273 		return (1);
1274 
1275 	switch (pd->af) {
1276 #ifdef INET
1277 	case AF_INET: {
1278 		struct ip *h = mtod(m, struct ip *);
1279 		src->scrub->pfss_ttl = h->ip_ttl;
1280 		break;
1281 	}
1282 #endif /* INET */
1283 #ifdef INET6
1284 	case AF_INET6: {
1285 		struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
1286 		src->scrub->pfss_ttl = h->ip6_hlim;
1287 		break;
1288 	}
1289 #endif /* INET6 */
1290 	}
1291 
1292 
1293 	/*
1294 	 * All normalizations below are only begun if we see the start of
1295 	 * the connections.  They must all set an enabled bit in pfss_flags
1296 	 */
1297 	if ((th->th_flags & TH_SYN) == 0)
1298 		return (0);
1299 
1300 
1301 	if (th->th_off > (sizeof(struct tcphdr) >> 2) && src->scrub &&
1302 	    pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
1303 		/* Diddle with TCP options */
1304 		int hlen;
1305 		opt = hdr + sizeof(struct tcphdr);
1306 		hlen = (th->th_off << 2) - sizeof(struct tcphdr);
1307 		while (hlen >= TCPOLEN_TIMESTAMP) {
1308 			switch (*opt) {
1309 			case TCPOPT_EOL:	/* FALLTHROUGH */
1310 			case TCPOPT_NOP:
1311 				opt++;
1312 				hlen--;
1313 				break;
1314 			case TCPOPT_TIMESTAMP:
1315 				if (opt[1] >= TCPOLEN_TIMESTAMP) {
1316 					src->scrub->pfss_flags |=
1317 					    PFSS_TIMESTAMP;
1318 					src->scrub->pfss_ts_mod =
1319 					    htonl(arc4random());
1320 
1321 					/* note PFSS_PAWS not set yet */
1322 					memcpy(&tsval, &opt[2],
1323 					    sizeof(u_int32_t));
1324 					memcpy(&tsecr, &opt[6],
1325 					    sizeof(u_int32_t));
1326 					src->scrub->pfss_tsval0 = ntohl(tsval);
1327 					src->scrub->pfss_tsval = ntohl(tsval);
1328 					src->scrub->pfss_tsecr = ntohl(tsecr);
1329 					getmicrouptime(&src->scrub->pfss_last);
1330 				}
1331 				/* FALLTHROUGH */
1332 			default:
1333 				hlen -= MAX(opt[1], 2);
1334 				opt += MAX(opt[1], 2);
1335 				break;
1336 			}
1337 		}
1338 	}
1339 
1340 	return (0);
1341 }
1342 
1343 void
1344 pf_normalize_tcp_cleanup(struct pf_state *state)
1345 {
1346 	if (state->src.scrub)
1347 		uma_zfree(V_pf_state_scrub_z, state->src.scrub);
1348 	if (state->dst.scrub)
1349 		uma_zfree(V_pf_state_scrub_z, state->dst.scrub);
1350 
1351 	/* Someday... flush the TCP segment reassembly descriptors. */
1352 }
1353 
1354 int
1355 pf_normalize_tcp_stateful(struct mbuf *m, int off, struct pf_pdesc *pd,
1356     u_short *reason, struct tcphdr *th, struct pf_state *state,
1357     struct pf_state_peer *src, struct pf_state_peer *dst, int *writeback)
1358 {
1359 	struct timeval uptime;
1360 	u_int32_t tsval, tsecr;
1361 	u_int tsval_from_last;
1362 	u_int8_t hdr[60];
1363 	u_int8_t *opt;
1364 	int copyback = 0;
1365 	int got_ts = 0;
1366 
1367 	KASSERT((src->scrub || dst->scrub),
1368 	    ("%s: src->scrub && dst->scrub!", __func__));
1369 
1370 	/*
1371 	 * Enforce the minimum TTL seen for this connection.  Negate a common
1372 	 * technique to evade an intrusion detection system and confuse
1373 	 * firewall state code.
1374 	 */
1375 	switch (pd->af) {
1376 #ifdef INET
1377 	case AF_INET: {
1378 		if (src->scrub) {
1379 			struct ip *h = mtod(m, struct ip *);
1380 			if (h->ip_ttl > src->scrub->pfss_ttl)
1381 				src->scrub->pfss_ttl = h->ip_ttl;
1382 			h->ip_ttl = src->scrub->pfss_ttl;
1383 		}
1384 		break;
1385 	}
1386 #endif /* INET */
1387 #ifdef INET6
1388 	case AF_INET6: {
1389 		if (src->scrub) {
1390 			struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
1391 			if (h->ip6_hlim > src->scrub->pfss_ttl)
1392 				src->scrub->pfss_ttl = h->ip6_hlim;
1393 			h->ip6_hlim = src->scrub->pfss_ttl;
1394 		}
1395 		break;
1396 	}
1397 #endif /* INET6 */
1398 	}
1399 
1400 	if (th->th_off > (sizeof(struct tcphdr) >> 2) &&
1401 	    ((src->scrub && (src->scrub->pfss_flags & PFSS_TIMESTAMP)) ||
1402 	    (dst->scrub && (dst->scrub->pfss_flags & PFSS_TIMESTAMP))) &&
1403 	    pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) {
1404 		/* Diddle with TCP options */
1405 		int hlen;
1406 		opt = hdr + sizeof(struct tcphdr);
1407 		hlen = (th->th_off << 2) - sizeof(struct tcphdr);
1408 		while (hlen >= TCPOLEN_TIMESTAMP) {
1409 			switch (*opt) {
1410 			case TCPOPT_EOL:	/* FALLTHROUGH */
1411 			case TCPOPT_NOP:
1412 				opt++;
1413 				hlen--;
1414 				break;
1415 			case TCPOPT_TIMESTAMP:
1416 				/* Modulate the timestamps.  Can be used for
1417 				 * NAT detection, OS uptime determination or
1418 				 * reboot detection.
1419 				 */
1420 
1421 				if (got_ts) {
1422 					/* Huh?  Multiple timestamps!? */
1423 					if (V_pf_status.debug >= PF_DEBUG_MISC) {
1424 						DPFPRINTF(("multiple TS??"));
1425 						pf_print_state(state);
1426 						printf("\n");
1427 					}
1428 					REASON_SET(reason, PFRES_TS);
1429 					return (PF_DROP);
1430 				}
1431 				if (opt[1] >= TCPOLEN_TIMESTAMP) {
1432 					memcpy(&tsval, &opt[2],
1433 					    sizeof(u_int32_t));
1434 					if (tsval && src->scrub &&
1435 					    (src->scrub->pfss_flags &
1436 					    PFSS_TIMESTAMP)) {
1437 						tsval = ntohl(tsval);
1438 						pf_change_proto_a(m, &opt[2],
1439 						    &th->th_sum,
1440 						    htonl(tsval +
1441 						    src->scrub->pfss_ts_mod),
1442 						    0);
1443 						copyback = 1;
1444 					}
1445 
1446 					/* Modulate TS reply iff valid (!0) */
1447 					memcpy(&tsecr, &opt[6],
1448 					    sizeof(u_int32_t));
1449 					if (tsecr && dst->scrub &&
1450 					    (dst->scrub->pfss_flags &
1451 					    PFSS_TIMESTAMP)) {
1452 						tsecr = ntohl(tsecr)
1453 						    - dst->scrub->pfss_ts_mod;
1454 						pf_change_proto_a(m, &opt[6],
1455 						    &th->th_sum, htonl(tsecr),
1456 						    0);
1457 						copyback = 1;
1458 					}
1459 					got_ts = 1;
1460 				}
1461 				/* FALLTHROUGH */
1462 			default:
1463 				hlen -= MAX(opt[1], 2);
1464 				opt += MAX(opt[1], 2);
1465 				break;
1466 			}
1467 		}
1468 		if (copyback) {
1469 			/* Copyback the options, caller copys back header */
1470 			*writeback = 1;
1471 			m_copyback(m, off + sizeof(struct tcphdr),
1472 			    (th->th_off << 2) - sizeof(struct tcphdr), hdr +
1473 			    sizeof(struct tcphdr));
1474 		}
1475 	}
1476 
1477 
1478 	/*
1479 	 * Must invalidate PAWS checks on connections idle for too long.
1480 	 * The fastest allowed timestamp clock is 1ms.  That turns out to
1481 	 * be about 24 days before it wraps.  XXX Right now our lowerbound
1482 	 * TS echo check only works for the first 12 days of a connection
1483 	 * when the TS has exhausted half its 32bit space
1484 	 */
1485 #define TS_MAX_IDLE	(24*24*60*60)
1486 #define TS_MAX_CONN	(12*24*60*60)	/* XXX remove when better tsecr check */
1487 
1488 	getmicrouptime(&uptime);
1489 	if (src->scrub && (src->scrub->pfss_flags & PFSS_PAWS) &&
1490 	    (uptime.tv_sec - src->scrub->pfss_last.tv_sec > TS_MAX_IDLE ||
1491 	    time_uptime - state->creation > TS_MAX_CONN))  {
1492 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
1493 			DPFPRINTF(("src idled out of PAWS\n"));
1494 			pf_print_state(state);
1495 			printf("\n");
1496 		}
1497 		src->scrub->pfss_flags = (src->scrub->pfss_flags & ~PFSS_PAWS)
1498 		    | PFSS_PAWS_IDLED;
1499 	}
1500 	if (dst->scrub && (dst->scrub->pfss_flags & PFSS_PAWS) &&
1501 	    uptime.tv_sec - dst->scrub->pfss_last.tv_sec > TS_MAX_IDLE) {
1502 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
1503 			DPFPRINTF(("dst idled out of PAWS\n"));
1504 			pf_print_state(state);
1505 			printf("\n");
1506 		}
1507 		dst->scrub->pfss_flags = (dst->scrub->pfss_flags & ~PFSS_PAWS)
1508 		    | PFSS_PAWS_IDLED;
1509 	}
1510 
1511 	if (got_ts && src->scrub && dst->scrub &&
1512 	    (src->scrub->pfss_flags & PFSS_PAWS) &&
1513 	    (dst->scrub->pfss_flags & PFSS_PAWS)) {
1514 		/* Validate that the timestamps are "in-window".
1515 		 * RFC1323 describes TCP Timestamp options that allow
1516 		 * measurement of RTT (round trip time) and PAWS
1517 		 * (protection against wrapped sequence numbers).  PAWS
1518 		 * gives us a set of rules for rejecting packets on
1519 		 * long fat pipes (packets that were somehow delayed
1520 		 * in transit longer than the time it took to send the
1521 		 * full TCP sequence space of 4Gb).  We can use these
1522 		 * rules and infer a few others that will let us treat
1523 		 * the 32bit timestamp and the 32bit echoed timestamp
1524 		 * as sequence numbers to prevent a blind attacker from
1525 		 * inserting packets into a connection.
1526 		 *
1527 		 * RFC1323 tells us:
1528 		 *  - The timestamp on this packet must be greater than
1529 		 *    or equal to the last value echoed by the other
1530 		 *    endpoint.  The RFC says those will be discarded
1531 		 *    since it is a dup that has already been acked.
1532 		 *    This gives us a lowerbound on the timestamp.
1533 		 *        timestamp >= other last echoed timestamp
1534 		 *  - The timestamp will be less than or equal to
1535 		 *    the last timestamp plus the time between the
1536 		 *    last packet and now.  The RFC defines the max
1537 		 *    clock rate as 1ms.  We will allow clocks to be
1538 		 *    up to 10% fast and will allow a total difference
1539 		 *    or 30 seconds due to a route change.  And this
1540 		 *    gives us an upperbound on the timestamp.
1541 		 *        timestamp <= last timestamp + max ticks
1542 		 *    We have to be careful here.  Windows will send an
1543 		 *    initial timestamp of zero and then initialize it
1544 		 *    to a random value after the 3whs; presumably to
1545 		 *    avoid a DoS by having to call an expensive RNG
1546 		 *    during a SYN flood.  Proof MS has at least one
1547 		 *    good security geek.
1548 		 *
1549 		 *  - The TCP timestamp option must also echo the other
1550 		 *    endpoints timestamp.  The timestamp echoed is the
1551 		 *    one carried on the earliest unacknowledged segment
1552 		 *    on the left edge of the sequence window.  The RFC
1553 		 *    states that the host will reject any echoed
1554 		 *    timestamps that were larger than any ever sent.
1555 		 *    This gives us an upperbound on the TS echo.
1556 		 *        tescr <= largest_tsval
1557 		 *  - The lowerbound on the TS echo is a little more
1558 		 *    tricky to determine.  The other endpoint's echoed
1559 		 *    values will not decrease.  But there may be
1560 		 *    network conditions that re-order packets and
1561 		 *    cause our view of them to decrease.  For now the
1562 		 *    only lowerbound we can safely determine is that
1563 		 *    the TS echo will never be less than the original
1564 		 *    TS.  XXX There is probably a better lowerbound.
1565 		 *    Remove TS_MAX_CONN with better lowerbound check.
1566 		 *        tescr >= other original TS
1567 		 *
1568 		 * It is also important to note that the fastest
1569 		 * timestamp clock of 1ms will wrap its 32bit space in
1570 		 * 24 days.  So we just disable TS checking after 24
1571 		 * days of idle time.  We actually must use a 12d
1572 		 * connection limit until we can come up with a better
1573 		 * lowerbound to the TS echo check.
1574 		 */
1575 		struct timeval delta_ts;
1576 		int ts_fudge;
1577 
1578 
1579 		/*
1580 		 * PFTM_TS_DIFF is how many seconds of leeway to allow
1581 		 * a host's timestamp.  This can happen if the previous
1582 		 * packet got delayed in transit for much longer than
1583 		 * this packet.
1584 		 */
1585 		if ((ts_fudge = state->rule.ptr->timeout[PFTM_TS_DIFF]) == 0)
1586 			ts_fudge = V_pf_default_rule.timeout[PFTM_TS_DIFF];
1587 
1588 		/* Calculate max ticks since the last timestamp */
1589 #define TS_MAXFREQ	1100		/* RFC max TS freq of 1Khz + 10% skew */
1590 #define TS_MICROSECS	1000000		/* microseconds per second */
1591 		delta_ts = uptime;
1592 		timevalsub(&delta_ts, &src->scrub->pfss_last);
1593 		tsval_from_last = (delta_ts.tv_sec + ts_fudge) * TS_MAXFREQ;
1594 		tsval_from_last += delta_ts.tv_usec / (TS_MICROSECS/TS_MAXFREQ);
1595 
1596 		if ((src->state >= TCPS_ESTABLISHED &&
1597 		    dst->state >= TCPS_ESTABLISHED) &&
1598 		    (SEQ_LT(tsval, dst->scrub->pfss_tsecr) ||
1599 		    SEQ_GT(tsval, src->scrub->pfss_tsval + tsval_from_last) ||
1600 		    (tsecr && (SEQ_GT(tsecr, dst->scrub->pfss_tsval) ||
1601 		    SEQ_LT(tsecr, dst->scrub->pfss_tsval0))))) {
1602 			/* Bad RFC1323 implementation or an insertion attack.
1603 			 *
1604 			 * - Solaris 2.6 and 2.7 are known to send another ACK
1605 			 *   after the FIN,FIN|ACK,ACK closing that carries
1606 			 *   an old timestamp.
1607 			 */
1608 
1609 			DPFPRINTF(("Timestamp failed %c%c%c%c\n",
1610 			    SEQ_LT(tsval, dst->scrub->pfss_tsecr) ? '0' : ' ',
1611 			    SEQ_GT(tsval, src->scrub->pfss_tsval +
1612 			    tsval_from_last) ? '1' : ' ',
1613 			    SEQ_GT(tsecr, dst->scrub->pfss_tsval) ? '2' : ' ',
1614 			    SEQ_LT(tsecr, dst->scrub->pfss_tsval0)? '3' : ' '));
1615 			DPFPRINTF((" tsval: %u  tsecr: %u  +ticks: %u  "
1616 			    "idle: %jus %lums\n",
1617 			    tsval, tsecr, tsval_from_last,
1618 			    (uintmax_t)delta_ts.tv_sec,
1619 			    delta_ts.tv_usec / 1000));
1620 			DPFPRINTF((" src->tsval: %u  tsecr: %u\n",
1621 			    src->scrub->pfss_tsval, src->scrub->pfss_tsecr));
1622 			DPFPRINTF((" dst->tsval: %u  tsecr: %u  tsval0: %u"
1623 			    "\n", dst->scrub->pfss_tsval,
1624 			    dst->scrub->pfss_tsecr, dst->scrub->pfss_tsval0));
1625 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
1626 				pf_print_state(state);
1627 				pf_print_flags(th->th_flags);
1628 				printf("\n");
1629 			}
1630 			REASON_SET(reason, PFRES_TS);
1631 			return (PF_DROP);
1632 		}
1633 
1634 		/* XXX I'd really like to require tsecr but it's optional */
1635 
1636 	} else if (!got_ts && (th->th_flags & TH_RST) == 0 &&
1637 	    ((src->state == TCPS_ESTABLISHED && dst->state == TCPS_ESTABLISHED)
1638 	    || pd->p_len > 0 || (th->th_flags & TH_SYN)) &&
1639 	    src->scrub && dst->scrub &&
1640 	    (src->scrub->pfss_flags & PFSS_PAWS) &&
1641 	    (dst->scrub->pfss_flags & PFSS_PAWS)) {
1642 		/* Didn't send a timestamp.  Timestamps aren't really useful
1643 		 * when:
1644 		 *  - connection opening or closing (often not even sent).
1645 		 *    but we must not let an attacker to put a FIN on a
1646 		 *    data packet to sneak it through our ESTABLISHED check.
1647 		 *  - on a TCP reset.  RFC suggests not even looking at TS.
1648 		 *  - on an empty ACK.  The TS will not be echoed so it will
1649 		 *    probably not help keep the RTT calculation in sync and
1650 		 *    there isn't as much danger when the sequence numbers
1651 		 *    got wrapped.  So some stacks don't include TS on empty
1652 		 *    ACKs :-(
1653 		 *
1654 		 * To minimize the disruption to mostly RFC1323 conformant
1655 		 * stacks, we will only require timestamps on data packets.
1656 		 *
1657 		 * And what do ya know, we cannot require timestamps on data
1658 		 * packets.  There appear to be devices that do legitimate
1659 		 * TCP connection hijacking.  There are HTTP devices that allow
1660 		 * a 3whs (with timestamps) and then buffer the HTTP request.
1661 		 * If the intermediate device has the HTTP response cache, it
1662 		 * will spoof the response but not bother timestamping its
1663 		 * packets.  So we can look for the presence of a timestamp in
1664 		 * the first data packet and if there, require it in all future
1665 		 * packets.
1666 		 */
1667 
1668 		if (pd->p_len > 0 && (src->scrub->pfss_flags & PFSS_DATA_TS)) {
1669 			/*
1670 			 * Hey!  Someone tried to sneak a packet in.  Or the
1671 			 * stack changed its RFC1323 behavior?!?!
1672 			 */
1673 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
1674 				DPFPRINTF(("Did not receive expected RFC1323 "
1675 				    "timestamp\n"));
1676 				pf_print_state(state);
1677 				pf_print_flags(th->th_flags);
1678 				printf("\n");
1679 			}
1680 			REASON_SET(reason, PFRES_TS);
1681 			return (PF_DROP);
1682 		}
1683 	}
1684 
1685 
1686 	/*
1687 	 * We will note if a host sends his data packets with or without
1688 	 * timestamps.  And require all data packets to contain a timestamp
1689 	 * if the first does.  PAWS implicitly requires that all data packets be
1690 	 * timestamped.  But I think there are middle-man devices that hijack
1691 	 * TCP streams immediately after the 3whs and don't timestamp their
1692 	 * packets (seen in a WWW accelerator or cache).
1693 	 */
1694 	if (pd->p_len > 0 && src->scrub && (src->scrub->pfss_flags &
1695 	    (PFSS_TIMESTAMP|PFSS_DATA_TS|PFSS_DATA_NOTS)) == PFSS_TIMESTAMP) {
1696 		if (got_ts)
1697 			src->scrub->pfss_flags |= PFSS_DATA_TS;
1698 		else {
1699 			src->scrub->pfss_flags |= PFSS_DATA_NOTS;
1700 			if (V_pf_status.debug >= PF_DEBUG_MISC && dst->scrub &&
1701 			    (dst->scrub->pfss_flags & PFSS_TIMESTAMP)) {
1702 				/* Don't warn if other host rejected RFC1323 */
1703 				DPFPRINTF(("Broken RFC1323 stack did not "
1704 				    "timestamp data packet. Disabled PAWS "
1705 				    "security.\n"));
1706 				pf_print_state(state);
1707 				pf_print_flags(th->th_flags);
1708 				printf("\n");
1709 			}
1710 		}
1711 	}
1712 
1713 
1714 	/*
1715 	 * Update PAWS values
1716 	 */
1717 	if (got_ts && src->scrub && PFSS_TIMESTAMP == (src->scrub->pfss_flags &
1718 	    (PFSS_PAWS_IDLED|PFSS_TIMESTAMP))) {
1719 		getmicrouptime(&src->scrub->pfss_last);
1720 		if (SEQ_GEQ(tsval, src->scrub->pfss_tsval) ||
1721 		    (src->scrub->pfss_flags & PFSS_PAWS) == 0)
1722 			src->scrub->pfss_tsval = tsval;
1723 
1724 		if (tsecr) {
1725 			if (SEQ_GEQ(tsecr, src->scrub->pfss_tsecr) ||
1726 			    (src->scrub->pfss_flags & PFSS_PAWS) == 0)
1727 				src->scrub->pfss_tsecr = tsecr;
1728 
1729 			if ((src->scrub->pfss_flags & PFSS_PAWS) == 0 &&
1730 			    (SEQ_LT(tsval, src->scrub->pfss_tsval0) ||
1731 			    src->scrub->pfss_tsval0 == 0)) {
1732 				/* tsval0 MUST be the lowest timestamp */
1733 				src->scrub->pfss_tsval0 = tsval;
1734 			}
1735 
1736 			/* Only fully initialized after a TS gets echoed */
1737 			if ((src->scrub->pfss_flags & PFSS_PAWS) == 0)
1738 				src->scrub->pfss_flags |= PFSS_PAWS;
1739 		}
1740 	}
1741 
1742 	/* I have a dream....  TCP segment reassembly.... */
1743 	return (0);
1744 }
1745 
1746 static int
1747 pf_normalize_tcpopt(struct pf_rule *r, struct mbuf *m, struct tcphdr *th,
1748     int off, sa_family_t af)
1749 {
1750 	u_int16_t	*mss;
1751 	int		 thoff;
1752 	int		 opt, cnt, optlen = 0;
1753 	int		 rewrite = 0;
1754 	u_char		 opts[TCP_MAXOLEN];
1755 	u_char		*optp = opts;
1756 
1757 	thoff = th->th_off << 2;
1758 	cnt = thoff - sizeof(struct tcphdr);
1759 
1760 	if (cnt > 0 && !pf_pull_hdr(m, off + sizeof(*th), opts, cnt,
1761 	    NULL, NULL, af))
1762 		return (rewrite);
1763 
1764 	for (; cnt > 0; cnt -= optlen, optp += optlen) {
1765 		opt = optp[0];
1766 		if (opt == TCPOPT_EOL)
1767 			break;
1768 		if (opt == TCPOPT_NOP)
1769 			optlen = 1;
1770 		else {
1771 			if (cnt < 2)
1772 				break;
1773 			optlen = optp[1];
1774 			if (optlen < 2 || optlen > cnt)
1775 				break;
1776 		}
1777 		switch (opt) {
1778 		case TCPOPT_MAXSEG:
1779 			mss = (u_int16_t *)(optp + 2);
1780 			if ((ntohs(*mss)) > r->max_mss) {
1781 				th->th_sum = pf_proto_cksum_fixup(m,
1782 				    th->th_sum, *mss, htons(r->max_mss), 0);
1783 				*mss = htons(r->max_mss);
1784 				rewrite = 1;
1785 			}
1786 			break;
1787 		default:
1788 			break;
1789 		}
1790 	}
1791 
1792 	if (rewrite)
1793 		m_copyback(m, off + sizeof(*th), thoff - sizeof(*th), opts);
1794 
1795 	return (rewrite);
1796 }
1797 
1798 #ifdef INET
1799 static void
1800 pf_scrub_ip(struct mbuf **m0, u_int32_t flags, u_int8_t min_ttl, u_int8_t tos)
1801 {
1802 	struct mbuf		*m = *m0;
1803 	struct ip		*h = mtod(m, struct ip *);
1804 
1805 	/* Clear IP_DF if no-df was requested */
1806 	if (flags & PFRULE_NODF && h->ip_off & htons(IP_DF)) {
1807 		u_int16_t ip_off = h->ip_off;
1808 
1809 		h->ip_off &= htons(~IP_DF);
1810 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0);
1811 	}
1812 
1813 	/* Enforce a minimum ttl, may cause endless packet loops */
1814 	if (min_ttl && h->ip_ttl < min_ttl) {
1815 		u_int16_t ip_ttl = h->ip_ttl;
1816 
1817 		h->ip_ttl = min_ttl;
1818 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_ttl, h->ip_ttl, 0);
1819 	}
1820 
1821 	/* Enforce tos */
1822 	if (flags & PFRULE_SET_TOS) {
1823 		u_int16_t	ov, nv;
1824 
1825 		ov = *(u_int16_t *)h;
1826 		h->ip_tos = tos | (h->ip_tos & IPTOS_ECN_MASK);
1827 		nv = *(u_int16_t *)h;
1828 
1829 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ov, nv, 0);
1830 	}
1831 
1832 	/* random-id, but not for fragments */
1833 	if (flags & PFRULE_RANDOMID && !(h->ip_off & ~htons(IP_DF))) {
1834 		uint16_t ip_id = h->ip_id;
1835 
1836 		ip_fillid(h);
1837 		h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_id, h->ip_id, 0);
1838 	}
1839 }
1840 #endif /* INET */
1841 
1842 #ifdef INET6
1843 static void
1844 pf_scrub_ip6(struct mbuf **m0, u_int8_t min_ttl)
1845 {
1846 	struct mbuf		*m = *m0;
1847 	struct ip6_hdr		*h = mtod(m, struct ip6_hdr *);
1848 
1849 	/* Enforce a minimum ttl, may cause endless packet loops */
1850 	if (min_ttl && h->ip6_hlim < min_ttl)
1851 		h->ip6_hlim = min_ttl;
1852 }
1853 #endif
1854